Blog · Review Essay · Last reviewed June 23, 2026

How We Became Posthuman and the Body Behind Information

N. Katherine Hayles's How We Became Posthuman is a book about the old dream that information can leave the body behind. For AI readers, its warning is immediate: intelligence should not be evaluated as pattern in the abstract. It is trained, carried, interpreted, embodied, funded, cooled, routed, narrated, and placed inside institutions that decide what a mind is allowed to be.

For this review, embodied information discipline means refusing to evaluate an AI system as a free-floating intelligence. Ask what material carrier, data history, labor process, interface, bodily trace, legal context, and institutional use give the information force.

The practical test is the body-to-action chain: what signal is taken from a body or social setting, how it is transformed into information, what inference is drawn, what decision or design action follows, and what recourse remains for the person whose life supplied the signal.

The Book

How We Became Posthuman: Virtual Bodies in Cybernetics, Literature, and Informatics was published by the University of Chicago Press in 1999. Chicago lists the book at 364 pages, with ISBNs 9780226321462 and 9780226321394, and describes it as a study of three linked stories: information becoming imagined as separable from its material carrier, the construction of the cyborg, and the dismantling of the liberal humanist subject in cybernetic discourse.

Hayles ranges from the postwar Macy Conferences on cybernetics to Norbert Wiener, second-wave cybernetics, artificial life, Philip K. Dick, Neal Stephenson, Richard Powers, and other literary treatments of machine intelligence and virtuality. Chicago's table of contents makes the structure plain: the book is not only about cyberculture, and not only about literary theory. It is an intellectual history of how a culture learned to imagine minds as informational patterns.

The book won the American Comparative Literature Association's Rene Wellek Prize. UCLA's faculty profile for Hayles also records the award and situates her work across literature, science, and technology. That matters less as a trophy than as a sign of the book's unusual range. It belongs with AI governance, media theory, and human-machine cognition because it asks what gets erased when intelligence is described as if bodies, environments, labor, and histories were secondary details.

Current Context

As of June 23, 2026, "posthuman" should be read carefully. It does not mean a completed successor species, a proven machine consciousness, or a prophecy of human replacement. In Hayles's argument it names a condition in which the boundaries among body, information, machine, institution, and self-understanding have been reorganized. The practical question is not whether the human vanishes. It is which parts of human life get recoded as information, who controls the recoding, and what forms of agency remain available afterward.

That question now reaches beyond classic cybernetics into ordinary AI infrastructure. Assistants, companions, wearables, biometric systems, workplace analytics, learning platforms, medical scribes, and neurotechnology all turn embodied traces into operational claims: fatigue, stress, attention, identity, productivity, risk, care need, intent, or trustworthiness. A model does not need consciousness to make those claims matter. It only needs a workflow that lets the claim shape access, treatment, price, work, care, or belief.

The current governance record reflects the same pressure. NIST says AI RMF 1.0 is being revised, while its Core still organizes risk work around govern, map, measure, and manage across the AI lifecycle; NIST's Generative AI Profile applies that lifecycle vocabulary to generative systems. UNESCO's 2025 neurotechnology recommendation, Colorado's biological-data law, California's neural-data amendment, and the EU AI Act's biometric, emotion-recognition, high-risk, and transparency provisions show the body returning as a policy object. The European Commission's current AI Act page says prohibited-practice and AI-literacy obligations entered application on February 2, 2025, GPAI obligations on August 2, 2025, Article 50 transparency rules in August 2026, and high-risk timelines are being reset through the 2026 AI omnibus political agreement to December 2, 2027 for specified high-risk areas and August 2, 2028 for systems embedded in regulated products.

Hayles gives the analytic reason those dates and categories matter. Once the body becomes information, privacy is only one part of the issue. The deeper issue is whether embodied life can be rendered actionable without becoming unanswerable: a signal becomes a feature, the feature becomes an inference, the inference enters a workflow, and the workflow begins to govern the person who supplied the signal.

How Information Lost Its Body

Hayles's central claim is not that information is fake. It is that information becomes dangerous when treated as if it were independent of the material systems that produce and carry it. A signal needs a medium. A calculation needs hardware. A model needs data, annotation, energy, architecture, tuning, interface, and use. A mind needs a body, even when that body is distributed across machines, institutions, and social practice.

This point cuts directly against the fantasy of upload without residue: the idea that a person could be reduced to a pattern, copied, transmitted, and restored without loss. Hayles is interested in why that fantasy became plausible. Cybernetics, information theory, computation, and science fiction did not merely describe new machines. They helped invent a new common sense in which bodies looked like containers and intelligence looked like transferable code.

For present AI, the same move appears whenever a model is treated as intelligence detached from its supply chain. The smooth answer hides the data labor. The benchmark hides the task design. The chatbot persona hides policy, reinforcement, and product incentives. The cloud interface hides the data center and the people being measured by the system.

The corrective is not to deny abstraction. Abstraction is necessary for science, engineering, law, and coordination. The danger begins when abstraction gets promoted from method to ontology: the representation stops being a useful reduction and becomes the thing institutions treat as more real than the people, places, and bodies it compressed.

A stronger embodied reading also keeps infrastructure in frame. Training data comes from authored, scraped, purchased, volunteered, surveilled, or annotated human activity. Inference runs on chips, power, cooling, networks, logs, safety policies, and user interfaces. The product returns through schools, clinics, workplaces, courts, homes, companion chats, and customer-service portals. A model's "body" is not metaphorical if the institution can point to the workers, sensors, accounts, devices, and records that let its output act.

Cybernetics and the Self

One strength of the book is its history of cybernetics as a theory of feedback, control, communication, and self-regulating systems. Hayles does not treat cybernetics as a neutral technical vocabulary. She follows how its concepts changed ideas of subjectivity: the human becomes a system exchanging information with other systems, a loop rather than a sealed interior.

That shift can be liberating. It breaks the fantasy of the autonomous, self-contained individual and makes room for distributed cognition, prosthesis, interdependence, and machine partnership. But it can also become a theory of managerial readability. If a person is a feedback system, then institutions will be tempted to measure, steer, optimize, and replace the loops they can observe.

That double edge is where Hayles remains useful. She refuses both simple panic and simple celebration. The posthuman is not just a monster story about machines replacing people, and not just a triumph story about minds escaping flesh. It is a contested condition in which embodiment has to be defended and reimagined rather than denied.

Fiction as Technical Memory

The literary chapters are not decoration. Hayles uses fiction as a record of cultural thinking about information. Bernard Wolfe, Philip K. Dick, Neal Stephenson, Richard Powers, Greg Bear, and others become diagnostic instruments: they show how technical ideas about feedback, virtuality, artificial life, and informatics enter stories about selfhood.

This is especially valuable for AI culture. Technical systems do not arrive alone. They arrive with myths: the uploaded mind, the helpful assistant, the emergent god, the rational optimizer, the companion who understands, the agent that acts for you. Fiction lets readers see those myths before they become product defaults.

Hayles's method also models a better way to read cyberculture. Do not ask only whether a fictional machine predicted a real machine. Ask what assumptions the story made available: which body it ignored, which agency it granted, which interface it trusted, which form of dependency it made feel inevitable.

The AI-Age Reading

Read in 2026, How We Became Posthuman is one of the better antidotes to disembodied AI talk.

Large models are often discussed as if they were free-floating minds. In practice they are embodied in training corpora, chip supply chains, evaluation regimes, data centers, interface patterns, company policies, terms of service, energy grids, user habits, and institutional workflows. The relevant "body" is not a silicon skull. It is the technical and organizational ecology that lets a pattern produce consequences.

That matters for accountability. If intelligence is framed as an abstract emergent property, responsibility can dissolve into awe. If intelligence is understood as embodied in systems, then governance has handles: data provenance, labor conditions, energy load, deployment context, user dependency, audit trails, appeal paths, and the right to refuse automated mediation.

The book also clarifies why AI companions are so powerful. A conversational model does not merely process text. It enters the user's feedback loop of memory, mood, interpretation, self-description, and social rehearsal. The danger is not that the machine has no body. The danger is that its working body is partly the user's life, partly the company's infrastructure, and partly the interface that makes that arrangement feel private. That is a safety problem even when the system has no inner experience.

The same reading applies to biometrics, wearables, consumer neurotechnology, workplace analytics, and learning systems. The body is not disappearing into information. It is being captured as information and returned as access, suspicion, optimization, personalization, therapy, identity, and evidence.

This is the recurring loop the site tracks in practical terms. The body becomes data; data becomes an interface; the interface becomes a workflow; the workflow becomes evidence about the body. A facial match changes a boarding gate, a stress score changes a manager's view, a companion memory changes self-description, a classroom attention metric changes instruction, and a neural feature may someday change what a product thinks the user intends.

Governance and Safety

As of June 23, 2026, Hayles's argument has a concrete governance edge. NIST's AI Risk Management Framework asks organizations to govern, map, measure, and manage AI risks across the lifecycle, and its Generative AI Profile treats governance, content provenance, pre-deployment testing, incident disclosure, privacy, environmental impact, human-AI configuration, and value-chain integration as practical risk concerns. That is embodied information discipline in institutional form: document the carrier, context, limits, and use conditions before treating output as authority.

Neurotechnology makes the point even sharper. UNESCO adopted a Recommendation on the Ethics of Neurotechnology in November 2025, warning that neurotechnology can measure, modulate, or stimulate the nervous system and that neural data can reveal sensitive information about thoughts, emotions, reactions, and identity. OECD's 2019 Recommendation on Responsible Innovation in Neurotechnology had already treated the field as an international governance problem. In the United States, Colorado's HB24-1058 became law in 2024 and expanded sensitive-data protections to biological data including neural data; California's SB 1223 added a consumer's neural data to the CCPA definition of sensitive personal information.

The EU AI Act adds a different boundary. Article 5 prohibits workplace and education emotion-inference systems except for medical or safety reasons, and prohibits biometric categorization systems that infer protected traits such as race, political opinions, trade-union membership, religious or philosophical beliefs, sex life, or sexual orientation. Article 50 also creates transparency duties for people interacting directly with AI systems and for certain synthetic outputs. The exact duties depend on jurisdiction and use case, but the direction is clear: bodily inference and interface ambiguity have become governance questions, not science-fiction motifs.

The policy lesson is not that every AI system is neurotechnology. It is that the boundary between information system and embodied life is increasingly thin. A model trained on voice, gait, face, stress, keystrokes, sleep, location, classroom behavior, workplace telemetry, or nervous-system data can turn bodily traces into administrative claims about identity, health, risk, productivity, attention, or trustworthiness.

The practical artifact should be an embodied-information inventory. It records the bodily or behavioral signal, the device or setting that captures it, the transformation from raw signal to feature or embedding, the inference claimed, the model and vendor involved, the human workflow affected, the retention period, the training or secondary-use rule, the deletion path, the review authority, and the recourse process. That inventory belongs beside data provenance, evaluation records, audit trails, data minimization, and algorithmic recourse.

The safeguards follow from the book's argument: minimize embodied data collection, name the context of consent, prohibit secondary use without new review, separate clinical or accessibility use from advertising and productivity monitoring, preserve data provenance, document model and interface limits, require human review for consequential decisions, keep appeal and deletion paths usable, and treat biometric, affective, behavioral, or neural inference as high-risk even when the interface feels ordinary.

An embodied-information review should ask concrete questions before deployment:

For companions and identity systems, the safety case should include attachment and self-description. Does the system encourage users to outsource memory, mood interpretation, spiritual judgment, health meaning, or personal identity to a privately governed interface? Does it store sensitive disclosures by default? Can users export, delete, and contest the profile that has been made of them? Hayles helps name why those are not only privacy questions. They are questions about which body the information is allowed to build.

Where the Book Needs Friction

The book is dense. Readers who want a direct AI policy manual will find literary criticism, cybernetic history, and theoretical argument braided together. That density is a strength, but it also means the argument sometimes moves through specialized debates rather than practical institutional questions.

Its late-1990s horizon also matters. Hayles wrote before social media platforms, transformer models, mobile cloud life, synthetic media pipelines, and AI agents became ordinary infrastructure. Some examples now feel period-specific. The deeper frame, however, has aged well because contemporary AI has intensified the very abstraction she analyzed: the tendency to treat information as if it could be separated from the material, social, and bodily conditions that make it meaningful.

There is also a productive tension in the book's defense of embodiment. A strong embodiment argument should not become biological nostalgia. Human cognition has always been extended by tools, language, institutions, writing, maps, rituals, databases, and machines. The task is not to retreat to a pure body before mediation. The task is to govern mediation without pretending that bodies no longer matter.

The book also needs sharper political economy than it always supplies. Embodiment is not only philosophical. It is built through supply chains, disability access, gendered care, racialized surveillance, data-center siting, content moderation, clinical trials, platform labor, and who gets to opt out without losing service, work, education, or care.

What This Changes

The practical lesson is embodied information discipline.

When a system claims intelligence, ask where the body is. What hardware carries it? What workers made its data legible? What categories shaped its world? What interface turns its outputs into authority? What institution profits from its apparent disembodiment? What human capacities are being extended, and which are being quietly rerouted through a vendor's machinery?

This is also a lesson about recursive reality. Once information is treated as cleaner than the world, institutions begin redesigning the world to fit the information. People adapt to forms, dashboards, prompts, scores, and machine-readable identities. The abstraction then returns as evidence that the model was right about what mattered.

The governance standard is therefore not only "keep a human in the loop." The human may already be the loop's data source, operator, subject, target, appeal officer, or liability shield. The stronger question is whether the person whose body or life made the information possible retains any practical power over the chain that turns that information into action.

Hayles gives AI readers a vocabulary for resisting that loop. Keep the body in the frame. Keep the medium visible. Keep the labor named. Keep the institution accountable. Intelligence is not less real because it is embodied. It becomes governable only when its embodiment is no longer hidden.

Source Discipline

This review separates Hayles's historical and theoretical argument from current governance claims. University of Chicago Press, ACLA, UCLA, Duke, and scholarly review records support the book, author, award, and reception context. NIST, UNESCO, OECD, Colorado, California, and EUR-Lex sources support the current governance context for AI lifecycle risk, generative-AI documentation, neurotechnology ethics, biometric and emotion-inference limits, and neural-data privacy.

Those sources do not all carry the same force. NIST and OECD guidance is not the same as binding law. UNESCO's recommendation is an international standard-setting instrument, not a domestic privacy statute. Colorado and California define neural or biological data differently. The EU AI Act provisions apply by jurisdiction, role, risk category, and phase-in date. The review therefore uses them as evidence that embodied inference is now a governance object, not as proof that one universal rule already governs every deployment.

The bounded claim is not that Hayles predicted every present AI product, or that AI systems are conscious, divine, or AGI. The claim is narrower: when intelligence is presented as detachable information, governance should look for the bodies, media, labor, data, energy, interfaces, and institutions that make the information act.

Sources

Book links are paid affiliate links. As an Amazon Associate I earn from qualifying purchases.

Buy on Amazon Browse Books

Return to Blog · Return to Books